What to do if my discriminant analysis has large misclassified number & percent?

[#1]

I have done Manova for my data analysis and found significant differences among my treatments but when I run separate univariate tests and try to find where the difference is; the results were non-significant… Some other forums said that the overall variables in MNOVA may contribute to the differences among the groups ( Overall effect of all the variables), but independently the variables may not contribute to the differences and suggested doing further discriminant analysis.

I have run a discriminant analysis with the 5 variables as the Y, Covariates, and the treatment groups as X, Categories. I have 46 data counts, but showed 17 as Numbers misclassified, and 36.95 for Percent misclassified. Is there any remedy to my data? Are these results valid or does a large misclassified number & percentage mean there are something wrong with my results?

[#2 · Accepted solution]

It is often the case with any model for a categorical response that the predictions exhibit a large misclassification rate despite significant predictors. The Canonical Plot, in your case, shows the overlap between the four groups, especially the two groups in the lower right of the plot.

So I do not think that there is anything wrong. It is just the limitation of the model based on these 5 covariates.